Hi Greg,
Upstream commit cf0d53ba4947 ("vfio/pci: Virtualize Maximum Read Request Size") and commit 523184972b28 ("vfio/pci: Virtualize Maximum Payload Size") fixes nasty PCIe virtualization issues for platforms that support Maximum Payload Size bigger than 128.
Issue shows up when a device is assigned to the guest machine as a passthrough. Guest machine configures the MPS/MRRS settings to values that are incompatible with the parent bridge device.
This causes PCIe transaction timeouts and AER errors to be spilled in the host kernel.
Please apply commit cf0d53ba4947 and 523184972b28 to all affected releases to fix the resulting regression.
Thanks, Sinan
On Thu, 19 Apr 2018 13:07:29 -0400 Sinan Kaya okaya@codeaurora.org wrote:
Hi Greg,
Upstream commit cf0d53ba4947 ("vfio/pci: Virtualize Maximum Read Request Size") and commit 523184972b28 ("vfio/pci: Virtualize Maximum Payload Size") fixes nasty PCIe virtualization issues for platforms that support Maximum Payload Size bigger than 128.
Issue shows up when a device is assigned to the guest machine as a passthrough. Guest machine configures the MPS/MRRS settings to values that are incompatible with the parent bridge device.
This causes PCIe transaction timeouts and AER errors to be spilled in the host kernel.
Please apply commit cf0d53ba4947 and 523184972b28 to all affected releases to fix the resulting regression.
Acked-by: Alex Williamson alex.williamson@redhat.com
On Thu, Apr 19, 2018 at 01:07:29PM -0400, Sinan Kaya wrote:
Hi Greg,
Upstream commit cf0d53ba4947 ("vfio/pci: Virtualize Maximum Read Request Size") and commit 523184972b28 ("vfio/pci: Virtualize Maximum Payload Size") fixes nasty PCIe virtualization issues for platforms that support Maximum Payload Size bigger than 128.
Issue shows up when a device is assigned to the guest machine as a passthrough. Guest machine configures the MPS/MRRS settings to values that are incompatible with the parent bridge device.
This causes PCIe transaction timeouts and AER errors to be spilled in the host kernel.
Please apply commit cf0d53ba4947 and 523184972b28 to all affected releases to fix the resulting regression.
The first patch does not apply to 4.4.y, and the second one does not apply anywhere :(
Can you provide a series of backported patches that apply for me to queue up?
thanks,
greg k-h
On 4/20/2018 2:55 AM, Greg KH wrote:
On Thu, Apr 19, 2018 at 01:07:29PM -0400, Sinan Kaya wrote:
Hi Greg,
Upstream commit cf0d53ba4947 ("vfio/pci: Virtualize Maximum Read Request Size") and commit 523184972b28 ("vfio/pci: Virtualize Maximum Payload Size") fixes nasty PCIe virtualization issues for platforms that support Maximum Payload Size bigger than 128.
Issue shows up when a device is assigned to the guest machine as a passthrough. Guest machine configures the MPS/MRRS settings to values that are incompatible with the parent bridge device.
This causes PCIe transaction timeouts and AER errors to be spilled in the host kernel.
Please apply commit cf0d53ba4947 and 523184972b28 to all affected releases to fix the resulting regression.
It looks like the commit 523184972b28 ("vfio/pci: Virtualize Maximum Payload Size") is already applied for 4.9.94.
Go ahead and apply commit cf0d53ba4947 ("vfio/pci: Virtualize Maximum Read Request Size") for 4.9.y.
[detached HEAD 2ee0ebb] vfio/pci: Virtualize Maximum Read Request Size Author: Alex Williamson alex.williamson@redhat.com Date: Mon Oct 2 12:39:10 2017 -0600 1 file changed, 26 insertions(+), 3 deletions(-)
I'll take a look at 4.4.y and check with Alex if these can be backported or not.
The first patch does not apply to 4.4.y, and the second one does not apply anywhere :(
Can you provide a series of backported patches that apply for me to queue up?
thanks,
greg k-h
On Fri, Apr 20, 2018 at 12:47:52PM -0400, Sinan Kaya wrote:
On 4/20/2018 2:55 AM, Greg KH wrote:
On Thu, Apr 19, 2018 at 01:07:29PM -0400, Sinan Kaya wrote:
Hi Greg,
Upstream commit cf0d53ba4947 ("vfio/pci: Virtualize Maximum Read Request Size") and commit 523184972b28 ("vfio/pci: Virtualize Maximum Payload Size") fixes nasty PCIe virtualization issues for platforms that support Maximum Payload Size bigger than 128.
Issue shows up when a device is assigned to the guest machine as a passthrough. Guest machine configures the MPS/MRRS settings to values that are incompatible with the parent bridge device.
This causes PCIe transaction timeouts and AER errors to be spilled in the host kernel.
Please apply commit cf0d53ba4947 and 523184972b28 to all affected releases to fix the resulting regression.
It looks like the commit 523184972b28 ("vfio/pci: Virtualize Maximum Payload Size") is already applied for 4.9.94.
Go ahead and apply commit cf0d53ba4947 ("vfio/pci: Virtualize Maximum Read Request Size") for 4.9.y.
Yes, added that to the 4.14.y and 4.9.y trees (you can't forget 4.14.y...)
thanks,
greg k-h
Hi Greg,
On 4/20/2018 12:47 PM, Sinan Kaya wrote:
I'll take a look at 4.4.y and check with Alex if these can be backported or not.
I worked with Alex today.
The commit 523184972b28 doesn't apply because of another missing patch on 4.4.y kernel.
commit ddf9dc0eb531 ("vfio-pci: Virtualize PCIe & AF FLR")
Please apply commit ddf9dc0eb531 ("vfio-pci: Virtualize PCIe & AF FLR") before applying 523184972b28.
commit ddf9dc0eb531 ("vfio-pci: Virtualize PCIe & AF FLR") also fixes a real issue as follows:
"This seems to fix several instance of devices getting into bad states with userspace drivers, like dpdk, running inside a VM."
It would be good to pull both to 4.y.
I'll have to work on cf0d53ba4947 next.
Sinan
On Fri, 20 Apr 2018 17:06:42 -0400 Sinan Kaya okaya@codeaurora.org wrote:
Hi Greg,
On 4/20/2018 12:47 PM, Sinan Kaya wrote:
I'll take a look at 4.4.y and check with Alex if these can be backported or not.
I worked with Alex today.
The commit 523184972b28 doesn't apply because of another missing patch on 4.4.y kernel.
commit ddf9dc0eb531 ("vfio-pci: Virtualize PCIe & AF FLR")
Please apply commit ddf9dc0eb531 ("vfio-pci: Virtualize PCIe & AF FLR") before applying 523184972b28.
commit ddf9dc0eb531 ("vfio-pci: Virtualize PCIe & AF FLR") also fixes a real issue as follows:
"This seems to fix several instance of devices getting into bad states with userspace drivers, like dpdk, running inside a VM."
It would be good to pull both to 4.y.
I'll have to work on cf0d53ba4947 next.
cf0d53ba4947 should be a clean pick to latest 4.4.y, after applying ddf9dc0eb531 followed by 523184972b28, at least that's the way it worked for me. The additional commit ddf9dc0eb531 made the v4.9 release, so it's already present in the other stable trees. Thanks,
Alex
On Fri, Apr 20, 2018 at 03:50:37PM -0600, Alex Williamson wrote:
On Fri, 20 Apr 2018 17:06:42 -0400 Sinan Kaya okaya@codeaurora.org wrote:
Hi Greg,
On 4/20/2018 12:47 PM, Sinan Kaya wrote:
I'll take a look at 4.4.y and check with Alex if these can be backported or not.
I worked with Alex today.
The commit 523184972b28 doesn't apply because of another missing patch on 4.4.y kernel.
commit ddf9dc0eb531 ("vfio-pci: Virtualize PCIe & AF FLR")
Please apply commit ddf9dc0eb531 ("vfio-pci: Virtualize PCIe & AF FLR") before applying 523184972b28.
commit ddf9dc0eb531 ("vfio-pci: Virtualize PCIe & AF FLR") also fixes a real issue as follows:
"This seems to fix several instance of devices getting into bad states with userspace drivers, like dpdk, running inside a VM."
It would be good to pull both to 4.y.
I'll have to work on cf0d53ba4947 next.
cf0d53ba4947 should be a clean pick to latest 4.4.y, after applying ddf9dc0eb531 followed by 523184972b28, at least that's the way it worked for me. The additional commit ddf9dc0eb531 made the v4.9 release, so it's already present in the other stable trees. Thanks,
Yes, that all seems to work properly now, everything should be queued up. If I've missed anything, please let me know.
thanks,
greg k-h
linux-stable-mirror@lists.linaro.org
-
Alex Williamson -
Greg KH -
Sinan Kaya